Nvidia: The Rival Broadcom Arms Its Customers AgainstNarrow moat

Broadcom (AVGO) — moat facet

Broadcom is paid by Nvidia's customers to reduce their dependence on Nvidia — so Nvidia attacked the one market where it could bundle, and took a fifth of Ethernet switching in a year.

Broadcom's relationship with Nvidia is the most consequential and least conventional rivalry in this collection. Broadcom does not sell a competing GPU. It co-designs the custom accelerators — XPUs — that Google, Meta, OpenAI and others build specifically so that they buy fewer Nvidia GPUs1. Broadcom is the armourer, paid by the customers to reduce their dependence on a supplier they cannot otherwise negotiate with.

Nvidia Ethernet switch share (%)<4%A year earlier21.5%2026Won by bundling networking with accelerators — not on silicon merit. Estimate.
Broadcom arms Nvidia's customers; Nvidia answered on the one front where it could bundle.

That is a comfortable position while it lasts, and it explains why Broadcom wins share as hyperscaler frustration with Nvidia's pricing grows. Nvidia's answer has been to attack Broadcom where Broadcom is strongest. Its Ethernet switch share went from under 4% to roughly 21.5% in one year2, won not on silicon merit but by bundling networking with the accelerators customers already had to buy — the same tactic Broadcom uses in software, applied in reverse.

The asymmetry that matters is customer alignment. Nvidia sells against its customers' interests when it prices GPUs; Broadcom sells with them. That buys real loyalty from the handful of firms that matter — and it is exactly why Nvidia can never quite dislodge Broadcom from XPU programs by out-engineering it.

Watch Ethernet switch share rather than accelerator share. Custom silicon is contested a generation at a time and moves slowly. Networking is where Nvidia has bundling leverage and where it has already demonstrated it can take a fifth of a market in twelve months. Nvidia's own page in this collection examines the same rivalry from the other side.

Moat trajectory: Holding steady

Both sides gained ground in different rooms. Broadcom's XPU business grew because hyperscaler discomfort with Nvidia grew; Nvidia took Ethernet switch share from under 4% to about 21.5% by bundling. Neither displaced the other where it is strong. The relationship is more balanced than a year ago, which for the smaller company is progress.

The number that tests this moat
Third-party estimate
Nvidia's Ethernet switch share
~21.5%, from under 4% in a year

Won by bundling networking with accelerators customers already had to buy, not on silicon merit — the same tactic Broadcom uses in software, applied against it. Custom accelerator share moves a generation at a time; networking is where Nvidia has leverage and has proved it can take a fifth of a market in twelve months.

Source: Third-party AI-networking market share analysis ↗
References
  1. Third-party estimateBroadcom's XPU customers include Google, Meta and OpenAI, building accelerators to reduce NVIDIA dependence.
    Tom's Hardware — the custom AI ASIC state of play (May 2026): Broadcom XPU customers incl. Google (long-term TPU agreement through 2031, Apr 2026), Meta, Anthropic, OpenAI and Apple; ~70% of the custom-accelerator design market; $60–90B FY2027 SAM guided; million-XPU clusters planned per customer — 2025–2026 · publ. May 2026 · source ↗
  2. Third-party estimateNVIDIA's Ethernet switch share went from under 4% to roughly 21.5% in one year by bundling with accelerators.
    Third-party AI-networking market analysis — NVIDIA's Ethernet switch share rose from under 4% to approximately 21.5% in a single year, won by bundling networking with the GPUs customers already required; Broadcom retains a near-monopoly position in hyperscaler Ethernet switch silicon with its Tomahawk and Trident families — 2026 · publ. 2026 · source ↗
Sources
Generated September 22, 2026